Comparison of models for stroke-free survival prediction in patients with CADASIL

Area of Science:

• Neurology
• Genetics
• Biostatistics

Background:

• Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL) is a genetic disorder caused by NOTCH3 gene mutations.
• CADASIL exhibits heterogeneous progression, impacting clinical scores and leading to various clinical events.
• Accurate prediction of disease progression, specifically stroke-free survival time, is essential for patient management.

Purpose of the Study:

• To compare the predictive performance of Cox proportional hazards regression with machine learning models for stroke-free survival in CADASIL.
• To evaluate the efficacy of four different feature selection approaches in conjunction with these models.
• To identify the optimal modeling strategy for assessing CADASIL disease progression.

Main Methods:

• Utilized demographic, clinical, and magnetic resonance imaging data from a cohort of 482 CADASIL patients.
• Employed Cox proportional hazards regression and machine learning models (gradient boosting, random survival forest).
• Applied a nested cross-validation procedure with LASSO feature selection, evaluating performance using time-dependent Brier Score and AUC at 5 years.

Main Results:

• The componentwise gradient boosting model with LASSO achieved the best overall performance (mean Brier score: 0.165).
• The random survival forest model with LASSO demonstrated the highest discrimination ability (mean AUC: 0.773).
• Both models, when combined with LASSO feature selection, outperformed traditional regression methods.

Conclusions:

• Machine learning models, especially gradient boosting and random survival forests, combined with LASSO feature selection, are effective for predicting stroke-free survival in CADASIL.
• These advanced modeling techniques offer improved accuracy and discrimination compared to traditional methods.
• The findings support the use of these predictive tools for better assessment and management of CADASIL patients.